GPU computing lab
CUDA in a
Linux tab.
Fly through a gyroid tunnel on your GPU. A CUDA program running inside this tab’s isolated Linux guest casts a ray for every pixel through WebGPU.
Move the camera, then switch to Ray steps to see where the kernel works hardest.
Build a kernel with Rust
Recorded preview
Booting Linux…640 × 360 · 230,400 rays
1 90 steps per ray
Rust + CUDA
Open playground ↗
Build the Mandelbrot renderer
Edit the Rust driver or its CUDA kernel, then compile with Cargo inside Linux.
CUDA shell
Run another kernel or type a command in the same guest that renders the tunnel.
Preparing Linux…
This guest has a marker in /tmp. Open a second tab and look for it there.
Compare in a new tab ↗
Inside the tab
The guest has an ARM64 CPU, a temporary filesystem and no internet connection. Your browser downloads the tools. The CUDA bridge translates supported kernels to WGSL and sends their work to WebGPU.
Credits & source
Built with unkernel, cudawg, Alpine Linux, DuckDB, xterm.js and CodeMirror. Licenses and source · View repository